The flexible job-shop scheduling problem (FJSP) is currently one of the most critical issues in process planning and manufacturing. The FJSP is studied with the goal of achieving the shortest makespan. Recently, some intelligent optimization algorithms have been applied to solve FJSP, but the key parameters of intelligent optimization algorithms cannot be dynamically adjusted during the solution process. Thus, the solutions cannot best meet the needs of production. To solve the problems of slow convergence speed and reaching a local optimum with the artificial bee colony (ABC) algorithm, an improved self-learning artificial bee colony algorithm (SLABC) based on reinforcement learning (RL) is proposed. In the SLABC algorithm, the number of updated dimensions of the ABC algorithm can be intelligently selected according to the RL algorithm, which improves the convergence speed and accuracy. In addition, a self-learning model of the SLABC algorithm is constructed and analyzed using Q-learning as the learning method of the algorithm, and the state determination and reward methods of the RL algorithm are designed and included in the environment of the artificial bee colony algorithm.Finally, this article verifies that SLABC has excellent convergence speed and accuracy in solving FJSP through Brandimarte instances.
K E Y W O R D Sartificial bee colony, flexible job-shop scheduling problem, reinforcement learning, self-learning artificial bee colony
INTRODUCTIONThe flexible job-shop scheduling problem (FJSP) is an extension of the classic job-shop scheduling problem, and it is a complex combinatorial optimization problem. 1 The FJSP has been a research hotspot over the years. In recent years, artificial intelligence optimization algorithms such as the ant colony optimization algorithm (ACO), 2,3 genetic algorithm (GA), 4-6 bee colony algorithm, [7][8][9][10][11] and various hybrid algorithms [12][13][14][15] have been usedto solve this problem, and some progress has been achieved, but a set of completely good solutions has not yet been reached; therefore, there is room for further research on this problem.Job-shop scheduling is a processing resource allocation problem. It reasonably arranges production resources, processing time, processing sequence, and so on, according to existing constraints to obtain the optimal cost or efficiency. 16 Due to the NP-hard characteristics of the FJSP, it is difficult to achieve global optimization even for small problems. Therefore, many researchers have begun to develop more effective solutions to obtain near-optimal solutions. Because of this trend, to solve the combinatorial optimization problem, many optimization algorithms have been developed. Wang et al. 17 proposed a random weighted hybrid particle swarm optimization algorithm (PSO) based on the second-order oscillation,